xref: /titanic_50/usr/src/uts/common/fs/nfs/nfs3_vfsops.c (revision 423be36caf64ae383e89ff6808a11c3039e6388e)
1 /*
2  * CDDL HEADER START
3  *
4  * The contents of this file are subject to the terms of the
5  * Common Development and Distribution License (the "License").
6  * You may not use this file except in compliance with the License.
7  *
8  * You can obtain a copy of the license at usr/src/OPENSOLARIS.LICENSE
9  * or http://www.opensolaris.org/os/licensing.
10  * See the License for the specific language governing permissions
11  * and limitations under the License.
12  *
13  * When distributing Covered Code, include this CDDL HEADER in each
14  * file and include the License file at usr/src/OPENSOLARIS.LICENSE.
15  * If applicable, add the following below this CDDL HEADER, with the
16  * fields enclosed by brackets "[]" replaced with your own identifying
17  * information: Portions Copyright [yyyy] [name of copyright owner]
18  *
19  * CDDL HEADER END
20  */
21 /*
22  * Copyright 2006 Sun Microsystems, Inc.  All rights reserved.
23  * Use is subject to license terms.
24  */
25 
26 /*
27  *	Copyright (c) 1983,1984,1985,1986,1987,1988,1989  AT&T.
28  *	All rights reserved.
29  */
30 
31 #pragma ident	"%Z%%M%	%I%	%E% SMI"
32 
33 #include <sys/param.h>
34 #include <sys/types.h>
35 #include <sys/systm.h>
36 #include <sys/cred.h>
37 #include <sys/vfs.h>
38 #include <sys/vnode.h>
39 #include <sys/pathname.h>
40 #include <sys/sysmacros.h>
41 #include <sys/kmem.h>
42 #include <sys/mkdev.h>
43 #include <sys/mount.h>
44 #include <sys/mntent.h>
45 #include <sys/statvfs.h>
46 #include <sys/errno.h>
47 #include <sys/debug.h>
48 #include <sys/cmn_err.h>
49 #include <sys/utsname.h>
50 #include <sys/bootconf.h>
51 #include <sys/modctl.h>
52 #include <sys/acl.h>
53 #include <sys/flock.h>
54 #include <sys/policy.h>
55 #include <sys/zone.h>
56 #include <sys/class.h>
57 #include <sys/socket.h>
58 #include <sys/netconfig.h>
59 #include <sys/tsol/tnet.h>
60 
61 #include <rpc/types.h>
62 #include <rpc/auth.h>
63 #include <rpc/clnt.h>
64 
65 #include <nfs/nfs.h>
66 #include <nfs/nfs_clnt.h>
67 #include <nfs/rnode.h>
68 #include <nfs/mount.h>
69 #include <nfs/nfs_acl.h>
70 
71 #include <fs/fs_subr.h>
72 
73 /*
74  * From rpcsec module (common/rpcsec).
75  */
76 extern int sec_clnt_loadinfo(struct sec_data *, struct sec_data **, model_t);
77 extern void sec_clnt_freeinfo(struct sec_data *);
78 
79 /*
80  * The order and contents of this structure must be kept in sync with that of
81  * rfsreqcnt_v3_tmpl in nfs_stats.c
82  */
83 static char *rfsnames_v3[] = {
84 	"null", "getattr", "setattr", "lookup", "access", "readlink", "read",
85 	"write", "create", "mkdir", "symlink", "mknod", "remove", "rmdir",
86 	"rename", "link", "readdir", "readdirplus", "fsstat", "fsinfo",
87 	"pathconf", "commit"
88 };
89 
90 /*
91  * This table maps from NFS protocol number into call type.
92  * Zero means a "Lookup" type call
93  * One  means a "Read" type call
94  * Two  means a "Write" type call
95  * This is used to select a default time-out.
96  */
97 static uchar_t call_type_v3[] = {
98 	0, 0, 1, 0, 0, 0, 1,
99 	2, 2, 2, 2, 2, 2, 2,
100 	2, 2, 1, 2, 0, 0, 0,
101 	2 };
102 
103 /*
104  * Similar table, but to determine which timer to use
105  * (only real reads and writes!)
106  */
107 static uchar_t timer_type_v3[] = {
108 	0, 0, 0, 0, 0, 0, 1,
109 	2, 0, 0, 0, 0, 0, 0,
110 	0, 0, 1, 1, 0, 0, 0,
111 	0 };
112 
113 /*
114  * This table maps from NFS protocol number into a call type
115  * for the semisoft mount option.
116  * Zero means do not repeat operation.
117  * One  means repeat.
118  */
119 static uchar_t ss_call_type_v3[] = {
120 	0, 0, 1, 0, 0, 0, 0,
121 	1, 1, 1, 1, 1, 1, 1,
122 	1, 1, 0, 0, 0, 0, 0,
123 	1 };
124 
125 /*
126  * nfs3 vfs operations.
127  */
128 static int	nfs3_mount(vfs_t *, vnode_t *, struct mounta *, cred_t *);
129 static int	nfs3_unmount(vfs_t *, int, cred_t *);
130 static int	nfs3_root(vfs_t *, vnode_t **);
131 static int	nfs3_statvfs(vfs_t *, struct statvfs64 *);
132 static int	nfs3_sync(vfs_t *, short, cred_t *);
133 static int	nfs3_vget(vfs_t *, vnode_t **, fid_t *);
134 static int	nfs3_mountroot(vfs_t *, whymountroot_t);
135 static void	nfs3_freevfs(vfs_t *);
136 
137 static int	nfs3rootvp(vnode_t **, vfs_t *, struct servinfo *,
138 		    int, cred_t *, zone_t *);
139 
140 /*
141  * Initialize the vfs structure
142  */
143 
144 static int nfs3fstyp;
145 vfsops_t *nfs3_vfsops;
146 
147 /*
148  * Debug variable to check for rdma based
149  * transport startup and cleanup. Controlled
150  * through /etc/system. Off by default.
151  */
152 extern int rdma_debug;
153 
154 int
155 nfs3init(int fstyp, char *name)
156 {
157 	static const fs_operation_def_t nfs3_vfsops_template[] = {
158 		VFSNAME_MOUNT, nfs3_mount,
159 		VFSNAME_UNMOUNT, nfs3_unmount,
160 		VFSNAME_ROOT, nfs3_root,
161 		VFSNAME_STATVFS, nfs3_statvfs,
162 		VFSNAME_SYNC, (fs_generic_func_p) nfs3_sync,
163 		VFSNAME_VGET, nfs3_vget,
164 		VFSNAME_MOUNTROOT, nfs3_mountroot,
165 		VFSNAME_FREEVFS, (fs_generic_func_p)nfs3_freevfs,
166 		NULL, NULL
167 	};
168 	int error;
169 
170 	error = vfs_setfsops(fstyp, nfs3_vfsops_template, &nfs3_vfsops);
171 	if (error != 0) {
172 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
173 		    "nfs3init: bad vfs ops template");
174 		return (error);
175 	}
176 
177 	error = vn_make_ops(name, nfs3_vnodeops_template, &nfs3_vnodeops);
178 	if (error != 0) {
179 		(void) vfs_freevfsops_by_type(fstyp);
180 		zcmn_err(GLOBAL_ZONEID, CE_WARN,
181 		    "nfs3init: bad vnode ops template");
182 		return (error);
183 	}
184 
185 	nfs3fstyp = fstyp;
186 
187 	return (0);
188 }
189 
190 void
191 nfs3fini(void)
192 {
193 }
194 
195 static void
196 nfs3_free_args(struct nfs_args *nargs, nfs_fhandle *fh)
197 {
198 
199 	if (fh)
200 		kmem_free(fh, sizeof (*fh));
201 
202 	if (nargs->knconf) {
203 		if (nargs->knconf->knc_protofmly)
204 			kmem_free(nargs->knconf->knc_protofmly,
205 				KNC_STRSIZE);
206 	if (nargs->knconf->knc_proto)
207 		kmem_free(nargs->knconf->knc_proto, KNC_STRSIZE);
208 		kmem_free(nargs->knconf, sizeof (*nargs->knconf));
209 		nargs->knconf = NULL;
210 	}
211 
212 	if (nargs->fh) {
213 		kmem_free(nargs->fh, strlen(nargs->fh) + 1);
214 		nargs->fh = NULL;
215 	}
216 
217 	if (nargs->hostname) {
218 		kmem_free(nargs->hostname, strlen(nargs->hostname) + 1);
219 		nargs->hostname = NULL;
220 	}
221 
222 	if (nargs->addr) {
223 		if (nargs->addr->buf) {
224 			ASSERT(nargs->addr->len);
225 			kmem_free(nargs->addr->buf, nargs->addr->len);
226 		}
227 		kmem_free(nargs->addr, sizeof (struct netbuf));
228 		nargs->addr = NULL;
229 	}
230 
231 	if (nargs->syncaddr) {
232 		ASSERT(nargs->syncaddr->len);
233 		if (nargs->syncaddr->buf) {
234 			ASSERT(nargs->syncaddr->len);
235 			kmem_free(nargs->syncaddr->buf, nargs->syncaddr->len);
236 		}
237 		kmem_free(nargs->syncaddr, sizeof (struct netbuf));
238 		nargs->syncaddr = NULL;
239 	}
240 
241 	if (nargs->netname) {
242 		kmem_free(nargs->netname, strlen(nargs->netname) + 1);
243 		nargs->netname = NULL;
244 	}
245 
246 	if (nargs->nfs_ext_u.nfs_extA.secdata) {
247 		sec_clnt_freeinfo(nargs->nfs_ext_u.nfs_extA.secdata);
248 		nargs->nfs_ext_u.nfs_extA.secdata = NULL;
249 	}
250 }
251 
252 static int
253 nfs3_copyin(char *data, int datalen, struct nfs_args *nargs, nfs_fhandle *fh)
254 {
255 
256 	int error;
257 	size_t nlen;			/* length of netname */
258 	size_t hlen;			/* length of hostname */
259 	char netname[MAXNETNAMELEN+1];  /* server's netname */
260 	struct netbuf addr;		/* server's address */
261 	struct netbuf syncaddr;		/* AUTH_DES time sync addr */
262 	struct knetconfig *knconf;	/* transport knetconfig structure */
263 	struct sec_data *secdata = NULL;	/* security data */
264 	STRUCT_DECL(nfs_args, args);    	/* nfs mount arguments */
265 	STRUCT_DECL(knetconfig, knconf_tmp);
266 	STRUCT_DECL(netbuf, addr_tmp);
267 	int flags;
268 	char *p, *pf;
269 	char *userbufptr;
270 
271 
272 	bzero(nargs, sizeof (*nargs));
273 
274 	STRUCT_INIT(args, get_udatamodel());
275 	bzero(STRUCT_BUF(args), SIZEOF_STRUCT(nfs_args, DATAMODEL_NATIVE));
276 	if (copyin(data, STRUCT_BUF(args), MIN(datalen, STRUCT_SIZE(args))))
277 		return (EFAULT);
278 
279 	nargs->wsize = STRUCT_FGET(args, wsize);
280 	nargs->rsize = STRUCT_FGET(args, rsize);
281 	nargs->timeo = STRUCT_FGET(args, timeo);
282 	nargs->retrans = STRUCT_FGET(args, retrans);
283 	nargs->acregmin = STRUCT_FGET(args, acregmin);
284 	nargs->acregmax = STRUCT_FGET(args, acregmax);
285 	nargs->acdirmin = STRUCT_FGET(args, acdirmin);
286 	nargs->acdirmax = STRUCT_FGET(args, acdirmax);
287 
288 	flags = STRUCT_FGET(args, flags);
289 	nargs->flags = flags;
290 
291 	addr.buf = NULL;
292 	syncaddr.buf = NULL;
293 
294 	/*
295 	 * Allocate space for a knetconfig structure and
296 	 * its strings and copy in from user-land.
297 	 */
298 	knconf = kmem_zalloc(sizeof (*knconf), KM_SLEEP);
299 	STRUCT_INIT(knconf_tmp, get_udatamodel());
300 	if (copyin(STRUCT_FGETP(args, knconf), STRUCT_BUF(knconf_tmp),
301 		STRUCT_SIZE(knconf_tmp))) {
302 		kmem_free(knconf, sizeof (*knconf));
303 		return (EFAULT);
304 	}
305 
306 	knconf->knc_semantics = STRUCT_FGET(knconf_tmp, knc_semantics);
307 	knconf->knc_protofmly = STRUCT_FGETP(knconf_tmp, knc_protofmly);
308 	knconf->knc_proto = STRUCT_FGETP(knconf_tmp, knc_proto);
309 	if (get_udatamodel() != DATAMODEL_LP64) {
310 		knconf->knc_rdev = expldev(STRUCT_FGET(knconf_tmp, knc_rdev));
311 	} else {
312 		knconf->knc_rdev = STRUCT_FGET(knconf_tmp, knc_rdev);
313 	}
314 
315 	pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
316 	p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
317 	error = copyinstr(knconf->knc_protofmly, pf, KNC_STRSIZE, NULL);
318 	if (error) {
319 		kmem_free(pf, KNC_STRSIZE);
320 		kmem_free(p, KNC_STRSIZE);
321 		kmem_free(knconf, sizeof (*knconf));
322 		return (error);
323 	}
324 
325 	error = copyinstr(knconf->knc_proto, p, KNC_STRSIZE, NULL);
326 	if (error) {
327 		kmem_free(pf, KNC_STRSIZE);
328 		kmem_free(p, KNC_STRSIZE);
329 		kmem_free(knconf, sizeof (*knconf));
330 		return (error);
331 	}
332 
333 
334 	knconf->knc_protofmly = pf;
335 	knconf->knc_proto = p;
336 
337 	nargs->knconf = knconf;
338 	/*
339 	 * Get server address
340 	 */
341 	STRUCT_INIT(addr_tmp, get_udatamodel());
342 	if (copyin(STRUCT_FGETP(args, addr), STRUCT_BUF(addr_tmp),
343 	STRUCT_SIZE(addr_tmp))) {
344 	error = EFAULT;
345 	goto errout;
346 	}
347 
348 	nargs->addr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP);
349 	userbufptr = STRUCT_FGETP(addr_tmp, buf);
350 	addr.len = STRUCT_FGET(addr_tmp, len);
351 	addr.buf = kmem_alloc(addr.len, KM_SLEEP);
352 	addr.maxlen = addr.len;
353 	if (copyin(userbufptr, addr.buf, addr.len)) {
354 		kmem_free(addr.buf, addr.len);
355 		error = EFAULT;
356 		goto errout;
357 	}
358 	bcopy(&addr, nargs->addr, sizeof (struct netbuf));
359 
360 	/*
361 	 * Get the root fhandle
362 	 */
363 
364 	if (copyin(STRUCT_FGETP(args, fh), fh, sizeof (nfs_fhandle))) {
365 		error = EFAULT;
366 		goto errout;
367 	}
368 
369 
370 	/*
371 	 * Get server's hostname
372 	 */
373 	if (flags & NFSMNT_HOSTNAME) {
374 		error = copyinstr(STRUCT_FGETP(args, hostname),
375 			netname, sizeof (netname), &hlen);
376 	if (error)
377 		goto errout;
378 	nargs->hostname = kmem_zalloc(hlen, KM_SLEEP);
379 	(void) strcpy(nargs->hostname, netname);
380 	} else {
381 	nargs->hostname = NULL;
382 	}
383 
384 
385 	/*
386 	 * If there are syncaddr and netname data, load them in. This is
387 	 * to support data needed for NFSV4 when AUTH_DH is the negotiated
388 	 * flavor via SECINFO. (instead of using MOUNT protocol in V3).
389 	 */
390 	netname[0] = '\0';
391 	if (flags & NFSMNT_SECURE) {
392 		if (STRUCT_FGETP(args, syncaddr) == NULL) {
393 			error = EINVAL;
394 			goto errout;
395 		}
396 		/* get syncaddr */
397 		STRUCT_INIT(addr_tmp, get_udatamodel());
398 		if (copyin(STRUCT_FGETP(args, syncaddr), STRUCT_BUF(addr_tmp),
399 			STRUCT_SIZE(addr_tmp))) {
400 			error = EINVAL;
401 			goto errout;
402 		}
403 		userbufptr = STRUCT_FGETP(addr_tmp, buf);
404 		syncaddr.len = STRUCT_FGET(addr_tmp, len);
405 		syncaddr.buf = kmem_alloc(syncaddr.len, KM_SLEEP);
406 		syncaddr.maxlen = syncaddr.len;
407 		if (copyin(userbufptr, syncaddr.buf, syncaddr.len)) {
408 			kmem_free(syncaddr.buf, syncaddr.len);
409 			error = EFAULT;
410 			goto errout;
411 		}
412 
413 		nargs->syncaddr = kmem_alloc(sizeof (struct netbuf), KM_SLEEP);
414 		bcopy(&syncaddr, nargs->syncaddr, sizeof (struct netbuf));
415 
416 		ASSERT(STRUCT_FGETP(args, netname));
417 
418 		if (copyinstr(STRUCT_FGETP(args, netname), netname,
419 			sizeof (netname), &nlen)) {
420 			error = EFAULT;
421 			goto errout;
422 		}
423 
424 		netname[nlen] = '\0';
425 		nargs->netname = kmem_zalloc(nlen, KM_SLEEP);
426 		(void) strcpy(nargs->netname, netname);
427 	}
428 
429 	/*
430 	 * Get the extention data which has the security data structure.
431 	 * This includes data for AUTH_SYS as well.
432 	 */
433 	if (flags & NFSMNT_NEWARGS) {
434 		nargs->nfs_args_ext = STRUCT_FGET(args, nfs_args_ext);
435 		if (nargs->nfs_args_ext == NFS_ARGS_EXTA ||
436 			nargs->nfs_args_ext == NFS_ARGS_EXTB) {
437 			/*
438 			 * Indicating the application is using the new
439 			 * sec_data structure to pass in the security
440 			 * data.
441 			 */
442 			if (STRUCT_FGETP(args,
443 				nfs_ext_u.nfs_extA.secdata) != NULL) {
444 				error = sec_clnt_loadinfo(
445 					(struct sec_data *)STRUCT_FGETP(args,
446 					nfs_ext_u.nfs_extA.secdata),
447 					&secdata, get_udatamodel());
448 			}
449 			nargs->nfs_ext_u.nfs_extA.secdata = secdata;
450 		}
451 	}
452 
453 	if (error)
454 		goto errout;
455 
456 	/*
457 	 * Failover support:
458 	 *
459 	 * We may have a linked list of nfs_args structures,
460 	 * which means the user is looking for failover.  If
461 	 * the mount is either not "read-only" or "soft",
462 	 * we want to bail out with EINVAL.
463 	 */
464 	if (nargs->nfs_args_ext == NFS_ARGS_EXTB)
465 		nargs->nfs_ext_u.nfs_extB.next =
466 			STRUCT_FGETP(args, nfs_ext_u.nfs_extB.next);
467 
468 errout:
469 	if (error)
470 		nfs3_free_args(nargs, fh);
471 
472 	return (error);
473 }
474 
475 
476 /*
477  * nfs mount vfsop
478  * Set up mount info record and attach it to vfs struct.
479  */
480 static int
481 nfs3_mount(vfs_t *vfsp, vnode_t *mvp, struct mounta *uap, cred_t *cr)
482 {
483 	struct nfs_args	*args = NULL;
484 	nfs_fhandle	*fhandle = NULL;
485 	char *data = uap->dataptr;
486 	int error;
487 	vnode_t *rtvp;			/* the server's root */
488 	mntinfo_t *mi;			/* mount info, pointed at by vfs */
489 	size_t nlen;			/* length of netname */
490 	struct knetconfig *knconf;	/* transport knetconfig structure */
491 	struct knetconfig *rdma_knconf;	/* rdma transport structure */
492 	rnode_t *rp;
493 	struct servinfo *svp;		/* nfs server info */
494 	struct servinfo *svp_tail = NULL; /* previous nfs server info */
495 	struct servinfo *svp_head;	/* first nfs server info */
496 	struct servinfo *svp_2ndlast;	/* 2nd last in server info list */
497 	struct sec_data *secdata;	/* security data */
498 	int flags, addr_type;
499 	zone_t *zone = nfs_zone();
500 	zone_t *mntzone = NULL;
501 
502 
503 	if ((error = secpolicy_fs_mount(cr, mvp, vfsp)) != 0)
504 		return (EPERM);
505 
506 	if (mvp->v_type != VDIR)
507 		return (ENOTDIR);
508 
509 	/*
510 	 * get arguments
511 	 *
512 	 * nfs_args is now versioned and is extensible, so
513 	 * uap->datalen might be different from sizeof (args)
514 	 * in a compatible situation.
515 	 */
516 
517 more:
518 
519 	if (!(uap->flags & MS_SYSSPACE)) {
520 		if (args == NULL)
521 			args = kmem_alloc(sizeof (struct nfs_args), KM_SLEEP);
522 		else {
523 			nfs3_free_args(args, fhandle);
524 			fhandle = NULL;
525 		}
526 		if (fhandle == NULL)
527 			fhandle = kmem_alloc(sizeof (nfs_fhandle), KM_SLEEP);
528 		error = nfs3_copyin(data, uap->datalen, args, fhandle);
529 		if (error) {
530 			if (args)
531 				kmem_free(args, sizeof (*args));
532 			return (error);
533 		}
534 	} else {
535 		args = (struct nfs_args *)data;
536 		fhandle = (nfs_fhandle *)args->fh;
537 	}
538 
539 
540 	flags = args->flags;
541 
542 	if (uap->flags & MS_REMOUNT) {
543 		size_t	n;
544 		char	name[FSTYPSZ];
545 
546 		if (uap->flags & MS_SYSSPACE) {
547 			error = copystr(uap->fstype, name, FSTYPSZ, &n);
548 		} else {
549 			nfs3_free_args(args, fhandle);
550 			kmem_free(args, sizeof (*args));
551 			error = copyinstr(uap->fstype, name, FSTYPSZ, &n);
552 		}
553 		if (error) {
554 			if (error == ENAMETOOLONG)
555 				return (EINVAL);
556 			return (error);
557 		}
558 
559 		/*
560 		 * This check is to ensure that the request is a
561 		 * genuine nfs remount request.
562 		 */
563 
564 		if (strncmp(name, "nfs", 3) != 0)
565 			return (EINVAL);
566 
567 		/*
568 		 * If the request changes the locking type, disallow the
569 		 * remount,
570 		 * because it's questionable whether we can transfer the
571 		 * locking state correctly.
572 		 */
573 
574 		if ((mi = VFTOMI(vfsp)) != NULL) {
575 			uint_t new_mi_llock;
576 			uint_t old_mi_llock;
577 
578 			new_mi_llock = (flags & NFSMNT_LLOCK) ? 1 : 0;
579 			old_mi_llock = (mi->mi_flags & MI_LLOCK) ? 1 : 0;
580 			if (old_mi_llock != new_mi_llock)
581 				return (EBUSY);
582 		}
583 		return (0);
584 	}
585 
586 	mutex_enter(&mvp->v_lock);
587 	if (!(uap->flags & MS_OVERLAY) &&
588 	    (mvp->v_count != 1 || (mvp->v_flag & VROOT))) {
589 		mutex_exit(&mvp->v_lock);
590 		if (!(uap->flags & MS_SYSSPACE)) {
591 			nfs3_free_args(args, fhandle);
592 			kmem_free(args, sizeof (*args));
593 		}
594 		return (EBUSY);
595 	}
596 	mutex_exit(&mvp->v_lock);
597 
598 	/* make sure things are zeroed for errout: */
599 	rtvp = NULL;
600 	mi = NULL;
601 	secdata = NULL;
602 
603 	/*
604 	 * A valid knetconfig structure is required.
605 	 */
606 	if (!(flags & NFSMNT_KNCONF)) {
607 		if (!(uap->flags & MS_SYSSPACE)) {
608 			nfs3_free_args(args, fhandle);
609 			kmem_free(args, sizeof (*args));
610 		}
611 		return (EINVAL);
612 	}
613 
614 	if ((strlen(args->knconf->knc_protofmly) >= KNC_STRSIZE) ||
615 		(strlen(args->knconf->knc_proto) >= KNC_STRSIZE)) {
616 		if (!(uap->flags & MS_SYSSPACE)) {
617 			nfs3_free_args(args, fhandle);
618 			kmem_free(args, sizeof (*args));
619 		}
620 		return (EINVAL);
621 	}
622 
623 	/*
624 	 * Allocate a servinfo struct.
625 	 */
626 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
627 	mutex_init(&svp->sv_lock, NULL, MUTEX_DEFAULT, NULL);
628 	if (svp_tail) {
629 		svp_2ndlast = svp_tail;
630 		svp_tail->sv_next = svp;
631 	} else {
632 		svp_head = svp;
633 		svp_2ndlast = svp;
634 	}
635 
636 	svp_tail = svp;
637 
638 	svp->sv_knconf = args->knconf;
639 	args->knconf = NULL;
640 
641 	if (args->addr == NULL || args->addr->buf == NULL) {
642 		error = EINVAL;
643 		goto errout;
644 	}
645 
646 	svp->sv_addr.maxlen = args->addr->maxlen;
647 	svp->sv_addr.len = args->addr->len;
648 	svp->sv_addr.buf = args->addr->buf;
649 	args->addr->buf = NULL;
650 
651 	/*
652 	 * Check the root fhandle length
653 	 */
654 	ASSERT(fhandle);
655 	if (fhandle->fh_len > NFS3_FHSIZE || fhandle->fh_len == 0) {
656 		error = EINVAL;
657 #ifdef DEBUG
658 		zcmn_err(getzoneid(), CE_WARN,
659 		    "nfs3_mount: got an invalid fhandle. fh_len = %d",
660 		    fhandle->fh_len);
661 		fhandle->fh_len = NFS_FHANDLE_LEN;
662 		nfs_printfhandle(fhandle);
663 #endif
664 		goto errout;
665 	}
666 
667 	bcopy(&fhandle->fh_buf, &svp->sv_fhandle.fh_buf, fhandle->fh_len);
668 	svp->sv_fhandle.fh_len = fhandle->fh_len;
669 
670 	/*
671 	 * Get server's hostname
672 	 */
673 	if (flags & NFSMNT_HOSTNAME) {
674 		if (args->hostname == NULL) {
675 			error = EINVAL;
676 			goto errout;
677 		}
678 		svp->sv_hostnamelen = strlen(args->hostname) + 1;
679 		svp->sv_hostname = args->hostname;
680 		args->hostname = NULL;
681 	} else {
682 		char *p = "unknown-host";
683 		svp->sv_hostnamelen = strlen(p) + 1;
684 		svp->sv_hostname = kmem_zalloc(svp->sv_hostnamelen, KM_SLEEP);
685 		(void) strcpy(svp->sv_hostname, p);
686 	}
687 
688 
689 	/*
690 	 * RDMA MOUNT SUPPORT FOR NFS v3:
691 	 * Establish, is it possible to use RDMA, if so overload the
692 	 * knconf with rdma specific knconf and free the orignal.
693 	 */
694 	if ((flags & NFSMNT_TRYRDMA) || (flags & NFSMNT_DORDMA)) {
695 		/*
696 		 * Determine the addr type for RDMA, IPv4 or v6.
697 		 */
698 		if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET) == 0)
699 			addr_type = AF_INET;
700 		else if (strcmp(svp->sv_knconf->knc_protofmly, NC_INET6) == 0)
701 			addr_type = AF_INET6;
702 
703 		if (rdma_reachable(addr_type, &svp->sv_addr,
704 			&rdma_knconf) == 0) {
705 			/*
706 			 * If successful, hijack the orignal knconf and
707 			 * replace with a new one, depending on the flags.
708 			 */
709 			svp->sv_origknconf = svp->sv_knconf;
710 			svp->sv_knconf = rdma_knconf;
711 			knconf = rdma_knconf;
712 		} else {
713 			if (flags & NFSMNT_TRYRDMA) {
714 #ifdef	DEBUG
715 				if (rdma_debug)
716 					zcmn_err(getzoneid(), CE_WARN,
717 					    "no RDMA onboard, revert\n");
718 #endif
719 			}
720 
721 			if (flags & NFSMNT_DORDMA) {
722 				/*
723 				 * If proto=rdma is specified and no RDMA
724 				 * path to this server is avialable then
725 				 * ditch this server.
726 				 * This is not included in the mountable
727 				 * server list or the replica list.
728 				 * Check if more servers are specified;
729 				 * Failover case, otherwise bail out of mount.
730 				 */
731 				if (args->nfs_args_ext ==
732 				    NFS_ARGS_EXTB &&
733 					args->nfs_ext_u.nfs_extB.next
734 					!= NULL) {
735 					data = (char *)
736 						args->nfs_ext_u.nfs_extB.next;
737 					if (uap->flags & MS_RDONLY &&
738 					    !(flags & NFSMNT_SOFT)) {
739 						if (svp_head->sv_next == NULL) {
740 							svp_tail = NULL;
741 							svp_2ndlast = NULL;
742 							sv_free(svp_head);
743 							goto more;
744 						} else {
745 							svp_tail = svp_2ndlast;
746 							svp_2ndlast->sv_next =
747 							    NULL;
748 							sv_free(svp);
749 							goto more;
750 						}
751 					}
752 				} else {
753 					/*
754 					 * This is the last server specified
755 					 * in the nfs_args list passed down
756 					 * and its not rdma capable.
757 					 */
758 					if (svp_head->sv_next == NULL) {
759 						/*
760 						 * Is this the only one
761 						 */
762 						error = EINVAL;
763 #ifdef	DEBUG
764 						if (rdma_debug)
765 							zcmn_err(getzoneid(),
766 							    CE_WARN,
767 							    "No RDMA srv");
768 #endif
769 						goto errout;
770 					} else {
771 						/*
772 						 * There is list, since some
773 						 * servers specified before
774 						 * this passed all requirements
775 						 */
776 						svp_tail = svp_2ndlast;
777 						svp_2ndlast->sv_next = NULL;
778 						sv_free(svp);
779 						goto proceed;
780 					}
781 				}
782 			}
783 		}
784 	}
785 
786 	/*
787 	 * Get the extention data which has the new security data structure.
788 	 */
789 	if (flags & NFSMNT_NEWARGS) {
790 		switch (args->nfs_args_ext) {
791 		case NFS_ARGS_EXTA:
792 		case NFS_ARGS_EXTB:
793 			/*
794 			 * Indicating the application is using the new
795 			 * sec_data structure to pass in the security
796 			 * data.
797 			 */
798 			secdata = args->nfs_ext_u.nfs_extA.secdata;
799 			if (args->nfs_ext_u.nfs_extA.secdata == NULL) {
800 				error = EINVAL;
801 			} else {
802 				/*
803 				 * Need to validate the flavor here if
804 				 * sysspace, userspace was already
805 				 * validate from the nfs_copyin function.
806 				 */
807 				switch (secdata->rpcflavor) {
808 				case AUTH_NONE:
809 				case AUTH_UNIX:
810 				case AUTH_LOOPBACK:
811 				case AUTH_DES:
812 				case RPCSEC_GSS:
813 					args->nfs_ext_u.nfs_extA.secdata =
814 						NULL;
815 					break;
816 				default:
817 					error = EINVAL;
818 					goto errout;
819 				}
820 			}
821 			break;
822 
823 		default:
824 			error = EINVAL;
825 			break;
826 		}
827 	} else if (flags & NFSMNT_SECURE) {
828 		/*
829 		 * Keep this for backward compatibility to support
830 		 * NFSMNT_SECURE/NFSMNT_RPCTIMESYNC flags.
831 		 */
832 		if (args->syncaddr == NULL || args->syncaddr->buf == NULL) {
833 			error = EINVAL;
834 			goto errout;
835 		}
836 		/*
837 		 * Move security related data to the sec_data structure.
838 		 */
839 		{
840 			dh_k4_clntdata_t *data;
841 			char *pf, *p;
842 			secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
843 			if (flags & NFSMNT_RPCTIMESYNC)
844 				secdata->flags |= AUTH_F_RPCTIMESYNC;
845 			data = kmem_alloc(sizeof (*data), KM_SLEEP);
846 			bcopy(args->syncaddr, &data->syncaddr,
847 				sizeof (*args->syncaddr));
848 
849 			/*
850 			 * duplicate the knconf information for the
851 			 * new opaque data.
852 			 */
853 			data->knconf = kmem_alloc(sizeof (*knconf), KM_SLEEP);
854 			*data->knconf = *knconf;
855 			pf = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
856 			p = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
857 			bcopy(knconf->knc_protofmly, pf, KNC_STRSIZE);
858 			bcopy(knconf->knc_proto, pf, KNC_STRSIZE);
859 			data->knconf->knc_protofmly = pf;
860 			data->knconf->knc_proto = p;
861 
862 			nlen = strlen(args->hostname) + 1;
863 			/* move server netname to the sec_data structure */
864 			if (nlen != 0) {
865 				data->netname = kmem_alloc(nlen, KM_SLEEP);
866 				bcopy(args->hostname, data->netname, nlen);
867 				data->netnamelen = nlen;
868 			}
869 			secdata->secmod = secdata->rpcflavor = AUTH_DES;
870 			secdata->data = (caddr_t)data;
871 		}
872 	} else 	{
873 		secdata = kmem_alloc(sizeof (*secdata), KM_SLEEP);
874 		secdata->secmod = secdata->rpcflavor = AUTH_UNIX;
875 		secdata->data = NULL;
876 	}
877 
878 	svp->sv_secdata = secdata;
879 	if (error)
880 		goto errout;
881 
882 	/*
883 	 * See bug 1180236.
884 	 * If mount secure failed, we will fall back to AUTH_NONE
885 	 * and try again.  nfs3rootvp() will turn this back off.
886 	 *
887 	 * The NFS Version 3 mount uses the FSINFO and GETATTR
888 	 * procedures.  The server should not care if these procedures
889 	 * have the proper security flavor, so if mount retries using
890 	 * AUTH_NONE that does not require a credential setup for root
891 	 * then the automounter would work without requiring root to be
892 	 * keylogged into AUTH_DES.
893 	 */
894 	if (secdata->rpcflavor != AUTH_UNIX &&
895 	    secdata->rpcflavor != AUTH_LOOPBACK)
896 		secdata->flags |= AUTH_F_TRYNONE;
897 
898 	/*
899 	 * Failover support:
900 	 *
901 	 * We may have a linked list of nfs_args structures,
902 	 * which means the user is looking for failover.  If
903 	 * the mount is either not "read-only" or "soft",
904 	 * we want to bail out with EINVAL.
905 	 */
906 	if (args->nfs_args_ext == NFS_ARGS_EXTB &&
907 	    args->nfs_ext_u.nfs_extB.next != NULL) {
908 		if (uap->flags & MS_RDONLY && !(flags & NFSMNT_SOFT)) {
909 			data = (char *)args->nfs_ext_u.nfs_extB.next;
910 			goto more;
911 		}
912 		error = EINVAL;
913 		goto errout;
914 	}
915 
916 	/*
917 	 * Determine the zone we're being mounted into.
918 	 */
919 	zone_hold(mntzone = zone);		/* start with this assumption */
920 	if (getzoneid() == GLOBAL_ZONEID) {
921 		zone_rele(mntzone);
922 		mntzone = zone_find_by_path(refstr_value(vfsp->vfs_mntpt));
923 		ASSERT(mntzone != NULL);
924 		if (mntzone != zone) {
925 			error = EBUSY;
926 			goto errout;
927 		}
928 	}
929 
930 	if (is_system_labeled()) {
931 		error = nfs_mount_label_policy(vfsp, &svp->sv_addr,
932 		    svp->sv_knconf, cr);
933 
934 		if (error > 0)
935 			goto errout;
936 
937 		if (error == -1) {
938 			/* change mount to read-only to prevent write-down */
939 			vfs_setmntopt(vfsp, MNTOPT_RO, NULL, 0);
940 		}
941 	}
942 
943 	/*
944 	 * Stop the mount from going any further if the zone is going away.
945 	 */
946 	if (zone_status_get(mntzone) >= ZONE_IS_SHUTTING_DOWN) {
947 		error = EBUSY;
948 		goto errout;
949 	}
950 
951 	/*
952 	 * Get root vnode.
953 	 */
954 proceed:
955 	error = nfs3rootvp(&rtvp, vfsp, svp_head, flags, cr, mntzone);
956 
957 	if (error)
958 		goto errout;
959 
960 	/*
961 	 * Set option fields in the mount info record
962 	 */
963 	mi = VTOMI(rtvp);
964 
965 	if (svp_head->sv_next)
966 		mi->mi_flags |= MI_LLOCK;
967 
968 	error = nfs_setopts(rtvp, DATAMODEL_NATIVE, args);
969 
970 errout:
971 	if (error) {
972 		if (rtvp != NULL) {
973 			rp = VTOR(rtvp);
974 			if (rp->r_flags & RHASHED)
975 				rp_rmhash(rp);
976 		}
977 		sv_free(svp_head);
978 		if (mi != NULL) {
979 			nfs_async_stop(vfsp);
980 			nfs_async_manager_stop(vfsp);
981 			if (mi->mi_io_kstats) {
982 				kstat_delete(mi->mi_io_kstats);
983 				mi->mi_io_kstats = NULL;
984 			}
985 			if (mi->mi_ro_kstats) {
986 				kstat_delete(mi->mi_ro_kstats);
987 				mi->mi_ro_kstats = NULL;
988 			}
989 			nfs_free_mi(mi);
990 		}
991 	}
992 
993 
994 	if (!(uap->flags & MS_SYSSPACE)) {
995 		nfs3_free_args(args, fhandle);
996 		kmem_free(args, sizeof (*args));
997 	}
998 
999 	if (rtvp != NULL)
1000 		VN_RELE(rtvp);
1001 
1002 	if (mntzone != NULL)
1003 		zone_rele(mntzone);
1004 
1005 	return (error);
1006 	}
1007 
1008 static int nfs3_dynamic = 0;	/* global variable to enable dynamic retrans. */
1009 static ushort_t nfs3_max_threads = 8;	/* max number of active async threads */
1010 static uint_t nfs3_bsize = 32 * 1024;	/* client `block' size */
1011 static uint_t nfs3_async_clusters = 1;	/* # of reqs from each async queue */
1012 static uint_t nfs3_cots_timeo = NFS_COTS_TIMEO;
1013 
1014 static int
1015 nfs3rootvp(vnode_t **rtvpp, vfs_t *vfsp, struct servinfo *svp,
1016 	int flags, cred_t *cr, zone_t *zone)
1017 {
1018 	vnode_t *rtvp;
1019 	mntinfo_t *mi;
1020 	dev_t nfs_dev;
1021 	struct vattr va;
1022 	struct FSINFO3args args;
1023 	struct FSINFO3res res;
1024 	int error;
1025 	int douprintf;
1026 	rnode_t *rp;
1027 	int i;
1028 	uint_t max_transfer_size;
1029 	struct nfs_stats *nfsstatsp;
1030 	cred_t *lcr = NULL, *tcr = cr;
1031 
1032 	nfsstatsp = zone_getspecific(nfsstat_zone_key, nfs_zone());
1033 	ASSERT(nfsstatsp != NULL);
1034 
1035 	ASSERT(nfs_zone() == zone);
1036 	/*
1037 	 * Create a mount record and link it to the vfs struct.
1038 	 */
1039 	mi = kmem_zalloc(sizeof (*mi), KM_SLEEP);
1040 	mutex_init(&mi->mi_lock, NULL, MUTEX_DEFAULT, NULL);
1041 	mutex_init(&mi->mi_remap_lock, NULL, MUTEX_DEFAULT, NULL);
1042 	mi->mi_flags = MI_ACL | MI_EXTATTR;
1043 	if (!(flags & NFSMNT_SOFT))
1044 		mi->mi_flags |= MI_HARD;
1045 	if ((flags & NFSMNT_SEMISOFT))
1046 		mi->mi_flags |= MI_SEMISOFT;
1047 	if ((flags & NFSMNT_NOPRINT))
1048 		mi->mi_flags |= MI_NOPRINT;
1049 	if (flags & NFSMNT_INT)
1050 		mi->mi_flags |= MI_INT;
1051 	mi->mi_retrans = NFS_RETRIES;
1052 	if (svp->sv_knconf->knc_semantics == NC_TPI_COTS_ORD ||
1053 	    svp->sv_knconf->knc_semantics == NC_TPI_COTS)
1054 		mi->mi_timeo = nfs3_cots_timeo;
1055 	else
1056 		mi->mi_timeo = NFS_TIMEO;
1057 	mi->mi_prog = NFS_PROGRAM;
1058 	mi->mi_vers = NFS_V3;
1059 	mi->mi_rfsnames = rfsnames_v3;
1060 	mi->mi_reqs = nfsstatsp->nfs_stats_v3.rfsreqcnt_ptr;
1061 	mi->mi_call_type = call_type_v3;
1062 	mi->mi_ss_call_type = ss_call_type_v3;
1063 	mi->mi_timer_type = timer_type_v3;
1064 	mi->mi_aclnames = aclnames_v3;
1065 	mi->mi_aclreqs = nfsstatsp->nfs_stats_v3.aclreqcnt_ptr;
1066 	mi->mi_acl_call_type = acl_call_type_v3;
1067 	mi->mi_acl_ss_call_type = acl_ss_call_type_v3;
1068 	mi->mi_acl_timer_type = acl_timer_type_v3;
1069 	cv_init(&mi->mi_failover_cv, NULL, CV_DEFAULT, NULL);
1070 	mi->mi_servers = svp;
1071 	mi->mi_curr_serv = svp;
1072 	mi->mi_acregmin = SEC2HR(ACREGMIN);
1073 	mi->mi_acregmax = SEC2HR(ACREGMAX);
1074 	mi->mi_acdirmin = SEC2HR(ACDIRMIN);
1075 	mi->mi_acdirmax = SEC2HR(ACDIRMAX);
1076 
1077 	if (nfs3_dynamic)
1078 		mi->mi_flags |= MI_DYNAMIC;
1079 
1080 	if (flags & NFSMNT_DIRECTIO)
1081 		mi->mi_flags |= MI_DIRECTIO;
1082 
1083 	/*
1084 	 * Make a vfs struct for nfs.  We do this here instead of below
1085 	 * because rtvp needs a vfs before we can do a getattr on it.
1086 	 *
1087 	 * Assign a unique device id to the mount
1088 	 */
1089 	mutex_enter(&nfs_minor_lock);
1090 	do {
1091 		nfs_minor = (nfs_minor + 1) & MAXMIN32;
1092 		nfs_dev = makedevice(nfs_major, nfs_minor);
1093 	} while (vfs_devismounted(nfs_dev));
1094 	mutex_exit(&nfs_minor_lock);
1095 
1096 	vfsp->vfs_dev = nfs_dev;
1097 	vfs_make_fsid(&vfsp->vfs_fsid, nfs_dev, nfs3fstyp);
1098 	vfsp->vfs_data = (caddr_t)mi;
1099 	vfsp->vfs_fstype = nfsfstyp;
1100 
1101 	/*
1102 	 * Verify that nfs3_bsize tuneable is set to an
1103 	 * acceptable value.  It be a multiple of PAGESIZE or
1104 	 * file corruption can occur.
1105 	 */
1106 	if (nfs3_bsize & PAGEOFFSET)
1107 		nfs3_bsize &= PAGEMASK;
1108 	if (nfs3_bsize < PAGESIZE)
1109 		nfs3_bsize = PAGESIZE;
1110 	vfsp->vfs_bsize = nfs3_bsize;
1111 
1112 	/*
1113 	 * Initialize fields used to support async putpage operations.
1114 	 */
1115 	for (i = 0; i < NFS_ASYNC_TYPES; i++)
1116 		mi->mi_async_clusters[i] = nfs3_async_clusters;
1117 	mi->mi_async_init_clusters = nfs3_async_clusters;
1118 	mi->mi_async_curr = &mi->mi_async_reqs[0];
1119 	mi->mi_max_threads = nfs3_max_threads;
1120 	mutex_init(&mi->mi_async_lock, NULL, MUTEX_DEFAULT, NULL);
1121 	cv_init(&mi->mi_async_reqs_cv, NULL, CV_DEFAULT, NULL);
1122 	cv_init(&mi->mi_async_work_cv, NULL, CV_DEFAULT, NULL);
1123 	cv_init(&mi->mi_async_cv, NULL, CV_DEFAULT, NULL);
1124 
1125 	mi->mi_vfsp = vfsp;
1126 	zone_hold(mi->mi_zone = zone);
1127 	nfs_mi_zonelist_add(mi);
1128 
1129 	/*
1130 	 * Make the root vnode, use it to get attributes,
1131 	 * then remake it with the attributes.
1132 	 */
1133 	rtvp = makenfs3node((nfs_fh3 *)&svp->sv_fhandle,
1134 	    NULL, vfsp, gethrtime(), cr, NULL, NULL);
1135 
1136 	/*
1137 	 * Make the FSINFO calls, primarily at this point to
1138 	 * determine the transfer size.  For client failover,
1139 	 * we'll want this to be the minimum bid from any
1140 	 * server, so that we don't overrun stated limits.
1141 	 *
1142 	 * While we're looping, we'll turn off AUTH_F_TRYNONE,
1143 	 * which is only for the mount operation.
1144 	 */
1145 
1146 	mi->mi_tsize = nfs3_tsize(svp->sv_knconf);
1147 	mi->mi_stsize = mi->mi_tsize;
1148 
1149 	mi->mi_curread = nfs3_bsize;
1150 	mi->mi_curwrite = mi->mi_curread;
1151 
1152 	/*
1153 	 * If the uid is set then set the creds for secure mounts
1154 	 * by proxy processes such as automountd.
1155 	 */
1156 	if (svp->sv_secdata->uid != 0 &&
1157 	    svp->sv_secdata->rpcflavor == RPCSEC_GSS) {
1158 		lcr = crdup(cr);
1159 		(void) crsetugid(lcr, svp->sv_secdata->uid, crgetgid(cr));
1160 		tcr = lcr;
1161 	}
1162 
1163 	for (svp = mi->mi_servers; svp != NULL; svp = svp->sv_next) {
1164 		douprintf = 1;
1165 		mi->mi_curr_serv = svp;
1166 		max_transfer_size = nfs3_tsize(svp->sv_knconf);
1167 		mi->mi_tsize = MIN(max_transfer_size, mi->mi_tsize);
1168 		mi->mi_stsize = MIN(max_transfer_size, mi->mi_stsize);
1169 		mi->mi_curread = MIN(max_transfer_size, mi->mi_curread);
1170 		mi->mi_curwrite = MIN(max_transfer_size, mi->mi_curwrite);
1171 		args.fsroot = *(nfs_fh3 *)&svp->sv_fhandle;
1172 
1173 		error = rfs3call(mi, NFSPROC3_FSINFO,
1174 		    xdr_nfs_fh3, (caddr_t)&args,
1175 		    xdr_FSINFO3res, (caddr_t)&res, tcr,
1176 		    &douprintf, &res.status, 0, NULL);
1177 		if (error)
1178 			goto bad;
1179 		error = geterrno3(res.status);
1180 		if (error)
1181 			goto bad;
1182 
1183 		/* get type of root node */
1184 		if (res.resok.obj_attributes.attributes) {
1185 			if (res.resok.obj_attributes.attr.type < NF3REG ||
1186 			    res.resok.obj_attributes.attr.type > NF3FIFO) {
1187 #ifdef DEBUG
1188 				zcmn_err(getzoneid(), CE_WARN,
1189 			    "NFS3 server %s returned a bad file type for root",
1190 				    svp->sv_hostname);
1191 #else
1192 				zcmn_err(getzoneid(), CE_WARN,
1193 			    "NFS server %s returned a bad file type for root",
1194 				    svp->sv_hostname);
1195 #endif
1196 				error = EINVAL;
1197 				goto bad;
1198 			} else {
1199 				if (rtvp->v_type != VNON &&
1200 		rtvp->v_type != nf3_to_vt[res.resok.obj_attributes.attr.type]) {
1201 #ifdef DEBUG
1202 					zcmn_err(getzoneid(), CE_WARN,
1203 		"NFS3 server %s returned a different file type for root",
1204 					    svp->sv_hostname);
1205 #else
1206 					zcmn_err(getzoneid(), CE_WARN,
1207 		"NFS server %s returned a different file type for root",
1208 					    svp->sv_hostname);
1209 #endif
1210 					error = EINVAL;
1211 					goto bad;
1212 				}
1213 				rtvp->v_type =
1214 				nf3_to_vt[res.resok.obj_attributes.attr.type];
1215 			}
1216 		}
1217 
1218 		if (res.resok.rtmax != 0) {
1219 			mi->mi_tsize = MIN(res.resok.rtmax, mi->mi_tsize);
1220 			if (res.resok.rtpref != 0) {
1221 				mi->mi_curread = MIN(res.resok.rtpref,
1222 						    mi->mi_curread);
1223 			} else {
1224 				mi->mi_curread = MIN(res.resok.rtmax,
1225 						    mi->mi_curread);
1226 			}
1227 		} else if (res.resok.rtpref != 0) {
1228 			mi->mi_tsize = MIN(res.resok.rtpref, mi->mi_tsize);
1229 			mi->mi_curread = MIN(res.resok.rtpref, mi->mi_curread);
1230 		} else {
1231 #ifdef DEBUG
1232 			zcmn_err(getzoneid(), CE_WARN,
1233 			    "NFS3 server %s returned 0 for read transfer sizes",
1234 			    svp->sv_hostname);
1235 #else
1236 			zcmn_err(getzoneid(), CE_WARN,
1237 			    "NFS server %s returned 0 for read transfer sizes",
1238 			    svp->sv_hostname);
1239 #endif
1240 			error = EIO;
1241 			goto bad;
1242 		}
1243 		if (res.resok.wtmax != 0) {
1244 			mi->mi_stsize = MIN(res.resok.wtmax, mi->mi_stsize);
1245 			if (res.resok.wtpref != 0) {
1246 				mi->mi_curwrite = MIN(res.resok.wtpref,
1247 						    mi->mi_curwrite);
1248 			} else {
1249 				mi->mi_curwrite = MIN(res.resok.wtmax,
1250 						    mi->mi_curwrite);
1251 			}
1252 		} else if (res.resok.wtpref != 0) {
1253 			mi->mi_stsize = MIN(res.resok.wtpref, mi->mi_stsize);
1254 			mi->mi_curwrite = MIN(res.resok.wtpref,
1255 					    mi->mi_curwrite);
1256 		} else {
1257 #ifdef DEBUG
1258 			zcmn_err(getzoneid(), CE_WARN,
1259 			"NFS3 server %s returned 0 for write transfer sizes",
1260 			    svp->sv_hostname);
1261 #else
1262 			zcmn_err(getzoneid(), CE_WARN,
1263 			"NFS server %s returned 0 for write transfer sizes",
1264 			    svp->sv_hostname);
1265 #endif
1266 			error = EIO;
1267 			goto bad;
1268 		}
1269 
1270 		/*
1271 		 * These signal the ability of the server to create
1272 		 * hard links and symbolic links, so they really
1273 		 * aren't relevant if there is more than one server.
1274 		 * We'll set them here, though it probably looks odd.
1275 		 */
1276 		if (res.resok.properties & FSF3_LINK)
1277 			mi->mi_flags |= MI_LINK;
1278 		if (res.resok.properties & FSF3_SYMLINK)
1279 			mi->mi_flags |= MI_SYMLINK;
1280 
1281 		/* Pick up smallest non-zero maxfilesize value */
1282 		if (res.resok.maxfilesize) {
1283 			if (mi->mi_maxfilesize) {
1284 				mi->mi_maxfilesize = MIN(mi->mi_maxfilesize,
1285 							res.resok.maxfilesize);
1286 			} else
1287 				mi->mi_maxfilesize = res.resok.maxfilesize;
1288 		}
1289 
1290 		/*
1291 		 * AUTH_F_TRYNONE is only for the mount operation,
1292 		 * so turn it back off.
1293 		 */
1294 		svp->sv_secdata->flags &= ~AUTH_F_TRYNONE;
1295 	}
1296 	mi->mi_curr_serv = mi->mi_servers;
1297 
1298 	/*
1299 	 * Start the thread responsible for handling async worker threads.
1300 	 */
1301 	VFS_HOLD(vfsp);	/* add reference for thread */
1302 	mi->mi_manager_thread = zthread_create(NULL, 0, nfs_async_manager,
1303 					vfsp, 0, minclsyspri);
1304 	ASSERT(mi->mi_manager_thread != NULL);
1305 
1306 	/*
1307 	 * Initialize kstats
1308 	 */
1309 	nfs_mnt_kstat_init(vfsp);
1310 
1311 	/* If we didn't get a type, get one now */
1312 	if (rtvp->v_type == VNON) {
1313 		va.va_mask = AT_ALL;
1314 
1315 		error = nfs3getattr(rtvp, &va, tcr);
1316 		if (error)
1317 			goto bad;
1318 		rtvp->v_type = va.va_type;
1319 	}
1320 
1321 	mi->mi_type = rtvp->v_type;
1322 
1323 	*rtvpp = rtvp;
1324 	if (lcr != NULL)
1325 		crfree(lcr);
1326 
1327 	return (0);
1328 bad:
1329 	/*
1330 	 * An error occurred somewhere, need to clean up...
1331 	 * We need to release our reference to the root vnode and
1332 	 * destroy the mntinfo struct that we just created.
1333 	 */
1334 	if (lcr != NULL)
1335 		crfree(lcr);
1336 	rp = VTOR(rtvp);
1337 	if (rp->r_flags & RHASHED)
1338 		rp_rmhash(rp);
1339 	VN_RELE(rtvp);
1340 	nfs_async_stop(vfsp);
1341 	nfs_async_manager_stop(vfsp);
1342 	if (mi->mi_io_kstats) {
1343 		kstat_delete(mi->mi_io_kstats);
1344 		mi->mi_io_kstats = NULL;
1345 	}
1346 	if (mi->mi_ro_kstats) {
1347 		kstat_delete(mi->mi_ro_kstats);
1348 		mi->mi_ro_kstats = NULL;
1349 	}
1350 	nfs_free_mi(mi);
1351 	*rtvpp = NULL;
1352 	return (error);
1353 }
1354 
1355 /*
1356  * vfs operations
1357  */
1358 static int
1359 nfs3_unmount(vfs_t *vfsp, int flag, cred_t *cr)
1360 {
1361 	mntinfo_t *mi;
1362 	ushort_t omax;
1363 
1364 	if (secpolicy_fs_unmount(cr, vfsp) != 0)
1365 		return (EPERM);
1366 
1367 	mi = VFTOMI(vfsp);
1368 	if (flag & MS_FORCE) {
1369 
1370 		vfsp->vfs_flag |= VFS_UNMOUNTED;
1371 
1372 		/*
1373 		 * We are about to stop the async manager.
1374 		 * Let every one know not to schedule any
1375 		 * more async requests
1376 		 */
1377 		mutex_enter(&mi->mi_async_lock);
1378 		mi->mi_max_threads = 0;
1379 		cv_broadcast(&mi->mi_async_work_cv);
1380 		mutex_exit(&mi->mi_async_lock);
1381 
1382 		/*
1383 		 * We need to stop the manager thread explicitly; the worker
1384 		 * threads can time out and exit on their own.
1385 		 */
1386 		nfs_async_manager_stop(vfsp);
1387 		destroy_rtable(vfsp, cr);
1388 		if (mi->mi_io_kstats) {
1389 			kstat_delete(mi->mi_io_kstats);
1390 			mi->mi_io_kstats = NULL;
1391 		}
1392 		if (mi->mi_ro_kstats) {
1393 			kstat_delete(mi->mi_ro_kstats);
1394 			mi->mi_ro_kstats = NULL;
1395 		}
1396 		return (0);
1397 	}
1398 	/*
1399 	 * Wait until all asynchronous putpage operations on
1400 	 * this file system are complete before flushing rnodes
1401 	 * from the cache.
1402 	 */
1403 	omax = mi->mi_max_threads;
1404 	if (nfs_async_stop_sig(vfsp)) {
1405 		return (EINTR);
1406 	}
1407 	rflush(vfsp, cr);
1408 	/*
1409 	 * If there are any active vnodes on this file system,
1410 	 * then the file system is busy and can't be umounted.
1411 	 */
1412 	if (check_rtable(vfsp)) {
1413 		mutex_enter(&mi->mi_async_lock);
1414 		mi->mi_max_threads = omax;
1415 		mutex_exit(&mi->mi_async_lock);
1416 		return (EBUSY);
1417 	}
1418 	/*
1419 	 * The unmount can't fail from now on; stop the worker thread manager.
1420 	 */
1421 	nfs_async_manager_stop(vfsp);
1422 	/*
1423 	 * Destroy all rnodes belonging to this file system from the
1424 	 * rnode hash queues and purge any resources allocated to
1425 	 * them.
1426 	 */
1427 	destroy_rtable(vfsp, cr);
1428 	if (mi->mi_io_kstats) {
1429 		kstat_delete(mi->mi_io_kstats);
1430 		mi->mi_io_kstats = NULL;
1431 	}
1432 	if (mi->mi_ro_kstats) {
1433 		kstat_delete(mi->mi_ro_kstats);
1434 		mi->mi_ro_kstats = NULL;
1435 	}
1436 	return (0);
1437 }
1438 
1439 /*
1440  * find root of nfs
1441  */
1442 static int
1443 nfs3_root(vfs_t *vfsp, vnode_t **vpp)
1444 {
1445 	mntinfo_t *mi;
1446 	vnode_t *vp;
1447 	servinfo_t *svp;
1448 
1449 	mi = VFTOMI(vfsp);
1450 
1451 	if (nfs_zone() != mi->mi_zone)
1452 		return (EPERM);
1453 
1454 	svp = mi->mi_curr_serv;
1455 	if (svp && (svp->sv_flags & SV_ROOT_STALE)) {
1456 		mutex_enter(&svp->sv_lock);
1457 		svp->sv_flags &= ~SV_ROOT_STALE;
1458 		mutex_exit(&svp->sv_lock);
1459 		return (ENOENT);
1460 	}
1461 
1462 	vp = makenfs3node((nfs_fh3 *)&mi->mi_curr_serv->sv_fhandle,
1463 	    NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1464 
1465 	if (VTOR(vp)->r_flags & RSTALE) {
1466 		VN_RELE(vp);
1467 		return (ENOENT);
1468 	}
1469 
1470 	ASSERT(vp->v_type == VNON || vp->v_type == mi->mi_type);
1471 
1472 	vp->v_type = mi->mi_type;
1473 
1474 	*vpp = vp;
1475 
1476 	return (0);
1477 }
1478 
1479 /*
1480  * Get file system statistics.
1481  */
1482 static int
1483 nfs3_statvfs(vfs_t *vfsp, struct statvfs64 *sbp)
1484 {
1485 	int error;
1486 	struct mntinfo *mi;
1487 	struct FSSTAT3args args;
1488 	struct FSSTAT3res res;
1489 	int douprintf;
1490 	failinfo_t fi;
1491 	vnode_t *vp;
1492 	cred_t *cr;
1493 	hrtime_t t;
1494 
1495 	mi = VFTOMI(vfsp);
1496 	if (nfs_zone() != mi->mi_zone)
1497 		return (EPERM);
1498 	error = nfs3_root(vfsp, &vp);
1499 	if (error)
1500 		return (error);
1501 
1502 	cr = CRED();
1503 
1504 	args.fsroot = *VTOFH3(vp);
1505 	fi.vp = vp;
1506 	fi.fhp = (caddr_t)&args.fsroot;
1507 	fi.copyproc = nfs3copyfh;
1508 	fi.lookupproc = nfs3lookup;
1509 	fi.xattrdirproc = acl_getxattrdir3;
1510 
1511 	douprintf = 1;
1512 
1513 	t = gethrtime();
1514 
1515 	error = rfs3call(mi, NFSPROC3_FSSTAT,
1516 	    xdr_nfs_fh3, (caddr_t)&args,
1517 	    xdr_FSSTAT3res, (caddr_t)&res, cr,
1518 	    &douprintf, &res.status, 0, &fi);
1519 
1520 	if (error) {
1521 		VN_RELE(vp);
1522 		return (error);
1523 	}
1524 
1525 	error = geterrno3(res.status);
1526 	if (!error) {
1527 		nfs3_cache_post_op_attr(vp, &res.resok.obj_attributes, t, cr);
1528 		sbp->f_bsize = MAXBSIZE;
1529 		sbp->f_frsize = DEV_BSIZE;
1530 		/*
1531 		 * Allow -1 fields to pass through unconverted.  These
1532 		 * indicate "don't know" fields.
1533 		 */
1534 		if (res.resok.tbytes == (size3)-1)
1535 			sbp->f_blocks = (fsblkcnt64_t)res.resok.tbytes;
1536 		else {
1537 			sbp->f_blocks = (fsblkcnt64_t)
1538 			    (res.resok.tbytes / DEV_BSIZE);
1539 		}
1540 		if (res.resok.fbytes == (size3)-1)
1541 			sbp->f_bfree = (fsblkcnt64_t)res.resok.fbytes;
1542 		else {
1543 			sbp->f_bfree = (fsblkcnt64_t)
1544 			    (res.resok.fbytes / DEV_BSIZE);
1545 		}
1546 		if (res.resok.abytes == (size3)-1)
1547 			sbp->f_bavail = (fsblkcnt64_t)res.resok.abytes;
1548 		else {
1549 			sbp->f_bavail = (fsblkcnt64_t)
1550 			    (res.resok.abytes / DEV_BSIZE);
1551 		}
1552 		sbp->f_files = (fsfilcnt64_t)res.resok.tfiles;
1553 		sbp->f_ffree = (fsfilcnt64_t)res.resok.ffiles;
1554 		sbp->f_favail = (fsfilcnt64_t)res.resok.afiles;
1555 		sbp->f_fsid = (unsigned long)vfsp->vfs_fsid.val[0];
1556 		(void) strncpy(sbp->f_basetype,
1557 		    vfssw[vfsp->vfs_fstype].vsw_name, FSTYPSZ);
1558 		sbp->f_flag = vf_to_stf(vfsp->vfs_flag);
1559 		sbp->f_namemax = (ulong_t)-1;
1560 	} else {
1561 		nfs3_cache_post_op_attr(vp, &res.resfail.obj_attributes, t, cr);
1562 		PURGE_STALE_FH(error, vp, cr);
1563 	}
1564 
1565 	VN_RELE(vp);
1566 
1567 	return (error);
1568 }
1569 
1570 static kmutex_t nfs3_syncbusy;
1571 
1572 /*
1573  * Flush dirty nfs files for file system vfsp.
1574  * If vfsp == NULL, all nfs files are flushed.
1575  */
1576 /* ARGSUSED */
1577 static int
1578 nfs3_sync(vfs_t *vfsp, short flag, cred_t *cr)
1579 {
1580 	/*
1581 	 * Cross-zone calls are OK here, since this translates to a
1582 	 * VOP_PUTPAGE(B_ASYNC), which gets picked up by the right zone.
1583 	 */
1584 	if (!(flag & SYNC_ATTR) && mutex_tryenter(&nfs3_syncbusy) != 0) {
1585 		rflush(vfsp, cr);
1586 		mutex_exit(&nfs3_syncbusy);
1587 	}
1588 	return (0);
1589 }
1590 
1591 /* ARGSUSED */
1592 static int
1593 nfs3_vget(vfs_t *vfsp, vnode_t **vpp, fid_t *fidp)
1594 {
1595 	int error;
1596 	nfs_fh3 fh;
1597 	vnode_t *vp;
1598 	struct vattr va;
1599 
1600 	if (fidp->fid_len > NFS3_FHSIZE) {
1601 		*vpp = NULL;
1602 		return (ESTALE);
1603 	}
1604 
1605 	if (nfs_zone() != VFTOMI(vfsp)->mi_zone)
1606 		return (EPERM);
1607 	fh.fh3_length = fidp->fid_len;
1608 	bcopy(fidp->fid_data, fh.fh3_u.data, fh.fh3_length);
1609 
1610 	vp = makenfs3node(&fh, NULL, vfsp, gethrtime(), CRED(), NULL, NULL);
1611 
1612 	if (VTOR(vp)->r_flags & RSTALE) {
1613 		VN_RELE(vp);
1614 		*vpp = NULL;
1615 		return (ENOENT);
1616 	}
1617 
1618 	if (vp->v_type == VNON) {
1619 		va.va_mask = AT_ALL;
1620 		error = nfs3getattr(vp, &va, CRED());
1621 		if (error) {
1622 			VN_RELE(vp);
1623 			*vpp = NULL;
1624 			return (error);
1625 		}
1626 		vp->v_type = va.va_type;
1627 	}
1628 
1629 	*vpp = vp;
1630 
1631 	return (0);
1632 }
1633 
1634 /* ARGSUSED */
1635 static int
1636 nfs3_mountroot(vfs_t *vfsp, whymountroot_t why)
1637 {
1638 	vnode_t *rtvp;
1639 	char root_hostname[SYS_NMLN+1];
1640 	struct servinfo *svp;
1641 	int error;
1642 	int vfsflags;
1643 	size_t size;
1644 	char *root_path;
1645 	struct pathname pn;
1646 	char *name;
1647 	cred_t *cr;
1648 	struct nfs_args args;		/* nfs mount arguments */
1649 	static char token[10];
1650 
1651 	bzero(&args, sizeof (args));
1652 
1653 	/* do this BEFORE getfile which causes xid stamps to be initialized */
1654 	clkset(-1L);		/* hack for now - until we get time svc? */
1655 
1656 	if (why == ROOT_REMOUNT) {
1657 		/*
1658 		 * Shouldn't happen.
1659 		 */
1660 		panic("nfs3_mountroot: why == ROOT_REMOUNT");
1661 	}
1662 
1663 	if (why == ROOT_UNMOUNT) {
1664 		/*
1665 		 * Nothing to do for NFS.
1666 		 */
1667 		return (0);
1668 	}
1669 
1670 	/*
1671 	 * why == ROOT_INIT
1672 	 */
1673 
1674 	name = token;
1675 	*name = 0;
1676 	getfsname("root", name, sizeof (token));
1677 
1678 	pn_alloc(&pn);
1679 	root_path = pn.pn_path;
1680 
1681 	svp = kmem_zalloc(sizeof (*svp), KM_SLEEP);
1682 	svp->sv_knconf = kmem_zalloc(sizeof (*svp->sv_knconf), KM_SLEEP);
1683 	svp->sv_knconf->knc_protofmly = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1684 	svp->sv_knconf->knc_proto = kmem_alloc(KNC_STRSIZE, KM_SLEEP);
1685 
1686 	/*
1687 	 * Get server address
1688 	 * Get the root fhandle
1689 	 * Get server's transport
1690 	 * Get server's hostname
1691 	 * Get options
1692 	 */
1693 	args.addr = &svp->sv_addr;
1694 	args.fh = (char *)&svp->sv_fhandle;
1695 	args.knconf = svp->sv_knconf;
1696 	args.hostname = root_hostname;
1697 	vfsflags = 0;
1698 	if (error = mount_root(*name ? name : "root", root_path, NFS_V3,
1699 	    &args, &vfsflags)) {
1700 		if (error == EPROTONOSUPPORT)
1701 			nfs_cmn_err(error, CE_WARN, "nfs3_mountroot: "
1702 			    "mount_root failed: server doesn't support NFS V3");
1703 		else
1704 			nfs_cmn_err(error, CE_WARN,
1705 			    "nfs3_mountroot: mount_root failed: %m");
1706 		sv_free(svp);
1707 		pn_free(&pn);
1708 		return (error);
1709 	}
1710 	svp->sv_hostnamelen = (int)(strlen(root_hostname) + 1);
1711 	svp->sv_hostname = kmem_alloc(svp->sv_hostnamelen, KM_SLEEP);
1712 	(void) strcpy(svp->sv_hostname, root_hostname);
1713 
1714 	/*
1715 	 * Force root partition to always be mounted with AUTH_UNIX for now
1716 	 */
1717 	svp->sv_secdata = kmem_alloc(sizeof (*svp->sv_secdata), KM_SLEEP);
1718 	svp->sv_secdata->secmod = AUTH_UNIX;
1719 	svp->sv_secdata->rpcflavor = AUTH_UNIX;
1720 	svp->sv_secdata->data = NULL;
1721 
1722 	cr = crgetcred();
1723 	rtvp = NULL;
1724 
1725 	error = nfs3rootvp(&rtvp, vfsp, svp, args.flags, cr, global_zone);
1726 
1727 	crfree(cr);
1728 
1729 	if (error) {
1730 		pn_free(&pn);
1731 		goto errout;
1732 	}
1733 
1734 	error = nfs_setopts(rtvp, DATAMODEL_NATIVE, &args);
1735 	if (error) {
1736 		nfs_cmn_err(error, CE_WARN,
1737 		    "nfs3_mountroot: invalid root mount options");
1738 		pn_free(&pn);
1739 		goto errout;
1740 	}
1741 
1742 	(void) vfs_lock_wait(vfsp);
1743 	vfs_add(NULL, vfsp, vfsflags);
1744 	vfs_unlock(vfsp);
1745 
1746 	size = strlen(svp->sv_hostname);
1747 	(void) strcpy(rootfs.bo_name, svp->sv_hostname);
1748 	rootfs.bo_name[size] = ':';
1749 	(void) strcpy(&rootfs.bo_name[size + 1], root_path);
1750 
1751 	pn_free(&pn);
1752 
1753 errout:
1754 	if (error) {
1755 		sv_free(svp);
1756 		nfs_async_stop(vfsp);
1757 		nfs_async_manager_stop(vfsp);
1758 	}
1759 
1760 	if (rtvp != NULL)
1761 		VN_RELE(rtvp);
1762 
1763 	return (error);
1764 }
1765 
1766 /*
1767  * Initialization routine for VFS routines.  Should only be called once
1768  */
1769 int
1770 nfs3_vfsinit(void)
1771 {
1772 	mutex_init(&nfs3_syncbusy, NULL, MUTEX_DEFAULT, NULL);
1773 	return (0);
1774 }
1775 
1776 void
1777 nfs3_vfsfini(void)
1778 {
1779 	mutex_destroy(&nfs3_syncbusy);
1780 }
1781 
1782 void
1783 nfs3_freevfs(vfs_t *vfsp)
1784 {
1785 	mntinfo_t *mi;
1786 	servinfo_t *svp;
1787 
1788 	/* free up the resources */
1789 	mi = VFTOMI(vfsp);
1790 	svp = mi->mi_servers;
1791 	mi->mi_servers = mi->mi_curr_serv = NULL;
1792 	sv_free(svp);
1793 
1794 	/*
1795 	 * By this time we should have already deleted the
1796 	 * mi kstats in the unmount code. If they are still around
1797 	 * somethings wrong
1798 	 */
1799 	ASSERT(mi->mi_io_kstats == NULL);
1800 	nfs_free_mi(mi);
1801 }
1802